Biology Atlas

Map / Concepts

concept · intermediate · partly written

How mutations change protein stability

A mutation changes the free energy of folding by an amount (ddG) that depends on what the residue was doing, and truncations act by a different mechanism than substitutions.

Also called: ddG, stability change on mutation, charge flip, truncation

The three rules of thumb, checked

Polar-for-hydrophobic in the core: supported, the biggest signal in the data. Charge flip: qualified, it depends on burial and on whether a partner exists. Truncation: right that it is worst, but the mechanism is loss of the protein (or a dominant truncated product), not a stability change. 12

Reported, not yet replicated

Substitutions more

The first principal component of substitution effects across 17,093 sites is hydrophobic preference (31% of variance), and most critical sites are core. A variance share is not an energy per mutation. At the surface, polar-for-hydrophobic is sometimes stabilising because the unfolded state also faces water. A charge flip at the surface changes an interaction worth under 1 kcal/mol; a buried charge can go either way (see the salt-bridge record). The median substitution destabilises because the wild type is usually more stable than a random substitute, and benchmark sets are skewed the same way. 13

Reported, not yet replicated

Truncations more

Per variant, truncating variants are about 26 to 27 times more often called pathogenic or likely pathogenic than missense (our arithmetic from the counts). Missense is still about one in five of all pathogenic records because there are about twenty times more of them. Two qualifications: a nonsense variant is not always a null (stops near the end of the gene escape nonsense-mediated decay and can make a stable truncated product, as in the HBB 3-prime end), and the classification itself is easier for truncations because of the default PVS1 criterion. 24

Our inference

Stability against function and conservation more

In 92 natural domains, positions that are strongly conserved yet have small average ddG are "functional sites": 7.5 to 9.3% of positions depending on definition, mostly on the surface at binding interfaces, holding 23% of the stabilising mutations. Stability predictors correlate with deep mutational scan scores at 0.26 on average. Use stability to propose a mechanism, not to rank pathogenicity. 56

Reported, not yet replicated

Interactions between mutations deep

Energies add but readouts bend: folded fraction is a sigmoid of dG, so a protein with spare stability shrugs off the first few mutations and then falls steeply. In protein G B1 about 4% of 509,693 double mutants show |epsilon| above 1. A double-mutant cycle isolates a specific coupling as ddG(double) - ddG(X) - ddG(Y); at scale, wild-type pair couplings are typically 0.5 to 1.0 kcal/mol, a few above 2. 781

Reported, not yet replicated

Limits of the dataset deep

Domains of 40 to 72 residues, in vitro, one pH and temperature, an assay window of about -1 to 5 kcal/mol, and partly folded states that can be cleaved and so underestimate stability. It says nothing direct about large multidomain proteins, binding, or function in cells. 1

Established

Plain answer

In the largest stability dataset (776,298 measurements on small domains) the median substitution destabilises by 0.59 kcal/mol and the strongest pattern is loss of buried hydrophobic groups (first principal component, 31% of variance); charge changes explain only 7% and depend on burial and partners. Truncation is a different mechanism: stop-gain and frameshift variants are called pathogenic in ClinVar far more often than missense, mostly through loss of the protein, not a stability change. 12

Reported, not yet replicated

Why it matters

Variant interpretation leans on three rules of thumb: a hydrophobic-to-polar change in the core is bad, a charge flip is bad, a truncation is worst. The data confirm the first, qualify the second and explain the third by a different mechanism, and they show that stability explains only a minority of what a variant does to function. 16

Reported, not yet replicated

Equation

ddG = dG(variant) - dG(wild type) ; fraction folded = 1 / (1 + exp(-dG / RT)) . Sign: Tsuboyama counts negative as destabilising; most older papers use the opposite sign.

Real numbers

Size and conditions of the mega-scale dataset

776,298 stability measurements (cDNA display proteolysis) on 331 natural and 148 designed domains of 40 to 72 residues, at pH 7.4, dynamic range about -1 to 5 kcal/mol; median ddG -0.59 kcal/mol; stabilising mutations (ddG above 1) are 0.2 to 0.6%. 1

Reported, not yet replicated

Axes of variation across 17,093 sites

Principal components: hydrophobic preference 31% of variance, helix-favouring 15%, aliphatic versus aromatic 12%, charge 7%, volume 6%. Burial is computed from AlphaFold models, so it is predicted. 1

Reported, not yet replicated

Cost of a buried Leu to Ala

About 3.2 +/- 1.2 kcal/mol in Pace's compilation (our arithmetic: at 298 K this lowers the folded fraction relative to unfolded roughly 220-fold, from 10^(3.2/1.364)). 3

Our inference

ClinVar pathogenic and likely pathogenic by consequence (live counts, 2026-11-17)

Nonsense 99,392 of 114,189 records (87.0%); frameshift 156,151 of 169,666 (92.0%); missense 87,605 of 2,606,011 (3.4%, with about 84% of missense records uncertain). The shares are our arithmetic from the counts; ascertainment is clinical testing and truncations get a default pathogenic criterion (PVS1), so 3.4% is a floor for missense, not a rate. 2

Our inference

Stops that escape nonsense-mediated decay

Only 51% of all possible stop-gains are predicted to trigger decay efficiently and 22% not; stops within 50 nt of the last exon junction, in the last exon, within 150 nt of the start, or in long exons typically escape. 4

Reported, not yet replicated

Population constraint

Median observed/expected for predicted loss-of-function variants 48%; the most constrained decile keeps about 6%; missense shows only modest depletion. 9

Reported, not yet replicated

Stability as a proxy for function

Across 49 deep mutational scan datasets, stability predictors correlate on average with Spearman 0.26 against functional scores; variant-effect predictors about 0.36. 6

Reported, not yet replicated

Teach this

In one sentence

Core hydrophobic loss is the strongest stability hit, charge flips depend on burial, and truncations break the protein by losing it rather than by destabilising it. 12

Reported, not yet replicated

The five-minute version

A mutation moves the folding free energy by ddG; one kcal/mol is about 1.7 thermal units, and every 1.36 kcal/mol moves the folded-to-unfolded ratio tenfold. The same ddG is lethal or silent depending on how much stability the protein had to spare and what the readout needs. Substitution rules of thumb hold in the core. Stop-gains are the worst class but act mainly by removing the protein or by the nonsense-mediated decay rules. 13

Reported, not yet replicated

What a listener will ask

Is a charge flip always bad?

No. At the surface a salt bridge is worth under 1 kcal/mol; buried charge is the dangerous case, and the answer depends on burial and partners. 3

Established

Are all stop-gains null alleles?

No. Premature stops in the last exon or near the last exon junction typically escape nonsense-mediated decay and can give a stable truncated protein.

Does ddG predict pathogenicity?

Only partly: stability predictors correlate with deep mutational scan scores at 0.26 on average. 6

Established

The common wrong idea

Any change of charge or hydrophobicity breaks a protein.

It depends on whether the residue is buried and what it was doing; most surface substitutions are close to neutral, and the median effect is small.

The one citation to hand over: 1

Still open here

Linked from

Sources

  1. Tsuboyama, K., Dauparas, J., Chen, J., et al.. Mega-scale experimental analysis of protein folding stability in biology and design. Nature (2023). paper; opened 2026-11-17 all records citing this
  2. ClinVar record counts by molecular consequence and clinical significance (NCBI E-utilities esearch). NCBI ClinVar (2026). database, not peer reviewed; opened 2026-11-17 all records citing this
  3. Pace, C. N., Scholtz, J. M., Grimsley, G. R.. Forces stabilizing proteins. FEBS Letters (2014). paper; opened 2026-11-17 all records citing this
  4. Lindeboom, R. G. H., Vermeulen, M., Lehner, B., Supek, F.. The impact of nonsense-mediated mRNA decay on genetic disease, gene editing and cancer immunotherapy. Nature Genetics (2019). paper; opened 2026-11-17 all records citing this
  5. Tsuboyama, K., et al.. Mega-scale experimental analysis of protein folding stability in biology and design (bioRxiv v3). bioRxiv (2022). preprint, not peer reviewed; opened 2026-11-17 all records citing this
  6. Gerasimavicius, L., Livesey, B. J., Marsh, J. A.. Correspondence between functional scores from deep mutational scans and predicted effects on protein stability. Protein Science (2023). paper; opened 2026-11-17 all records citing this
  7. Olson, C. A., Wu, N. C., Sun, R.. A comprehensive biophysical description of pairwise epistasis throughout an entire protein domain. Current Biology (2014). paper; opened 2026-11-17 all records citing this
  8. Pagano, L., Toto, A., Malagrino, F., et al.. Double mutant cycles as a tool to address folding, binding, and allostery. International Journal of Molecular Sciences (2021). paper; opened 2026-11-17 all records citing this
  9. Karczewski, K. J., Francioli, L. C., Tiao, G., et al.. The mutational constraint spectrum quantified from variation in 141,456 humans. Nature (2020). paper; opened 2026-11-17 all records citing this
History of this record (1)

This record has been changed 1 time, on the dates below. The note kept against each change is written for the people who maintain the atlas, in their own shorthand, so it is not shown here; it is in the record itself.

Status
partly written
Last updated
2026-11-17
Record
content/concepts/mutation-effects-on-protein-stability.json